Here is the brief. A person is standing in a car park. They are holding a phone in one hand and car keys in the other. It is bright, they are in a hurry, and they are not a photographer.

You need eight specific photos from them, from specific angles, consistent enough that all eight look like they belong to the same set.

Most capture flows fail at this, and they fail in the same few ways. These are the lessons that came out of building one.

## Nobody reads the instructions

The instinct is to explain. "Stand about three metres from the front left corner of the vehicle, with the car filling roughly 80 percent of the frame."

Nobody reads that. Not because users are careless, but because they are standing outside holding things, and reading is a task you just added on top of the task they came to do.

Replace the sentence with a picture. A faint outline on the camera view showing where the car should sit does the entire job with no reading at all. People align things to shapes automatically. They do not parse spatial instructions.

If you have text at all, it should be three words. "Front left corner."

Show the target, not the error

Telling someone they got it wrong after the fact is the worst possible moment. They have already moved, already tapped, already mentally finished that step.

Put the guidance before the shutter. An outline they are aligning to, that changes colour when they are close enough, is a conversation that happens while they can still act on it. The user fixes the problem without ever being told there was one.

This is the single biggest difference between a flow people finish and a flow people abandon.

## Validate immediately, but never block

Check the photo the instant it is taken. Too dark, too blurry, object not found, obviously the wrong angle. Those checks are cheap and they catch most real problems.

But do not trap people. If someone has tried twice and it is still not passing, let them through. Mark the photo as needing review and move on.

A flow that refuses to advance is a flow that gets closed. You will lose the other six photos as well as the one you were arguing about, and you will never get that person back to the car park.

Accept and flag beats reject and insist, every time.

## Progress is the motivator, not quality

A counter showing "3 of 8" does more for completion than any amount of feedback about image quality.

People finish things that look finishable. A visible end point turns an open-ended chore into a short list, and short lists get completed. Quality nagging does the opposite, because every correction makes the end feel further away.

Show progress prominently. Show quality feedback quietly.

## Assume the worst conditions, because they are the normal ones

Direct sunlight means they cannot see your subtle grey overlay. Use high contrast.

One hand is busy. Keep everything reachable with a thumb, and never put a critical control at the top of the screen.

Signal is bad. Car parks are famously bad for this, and often half underground. Capture locally, queue the upload, and let them walk away. If your flow requires a successful upload before the next photo, you have built something that fails outdoors, which is the only place it will ever be used.

And assume they will be interrupted. Save state after every single photo so that a phone call does not cost them the whole session.

## The thing that is actually hard

None of the above is technically difficult. The hard part is accepting that the person using your flow does not care about it and never will.

They are not evaluating your interface. They want to put the keys down and go back inside. Every decision gets easier once you design for that person instead of the attentive user in your head.

We wrote up what good capture actually looks like from the photography side, which is the other half of this problem.